Installation:

The ggplot2 package is a part of the tidyverse package. If you have the tidyverse package you don’t have to reinstall ggplot2 but you can always reinstall individually by using the following code.

# The easiest way to get ggplot2 is to install the whole tidyverse:
install.packages("tidyverse")
library(tidyverse)

# Alternatively, install just ggplot2:
install.packages("ggplot2")

library(ggplot2) # to load the package for every session, needs to be run every time you start a session.

\(\cdot\) geom_point():

The point geom is used to create scatterplots. The scatterplot is most useful for displaying the relationship between two continuous variables and thus one of the most commonly used geom functions. It can be used to compare one continuous and one categorical variable, or two categorical variables

The most commonly used aesthetics of geom_point() are the size, shape and color. More about the size, shape and color arguments can be found here.

Example:

ggplot(mpg, aes(x = cty, y = hwy)) +
  geom_point(color = "purple", size = 4, shape = 24, fill = "yellow")

Here, you can see that we used the both color and fill for the points this is because the shape used for the points is a hollow shape. Most of the shapes for points are solid points and just use color as aesthetic.

\(\cdot\) geom_line():

The line geom is used to create connected scatterplots. geom_line() connects the points in the order of the variable on the x axis. An equivalent function is the geom_path() which connects the observations in the order in which they appear in the data. Just like geom_point, geom_line is one of the most commonly used geoms and often used in combination with geom_point. In this course we will only work with geom_line.

The most commonly used aesthetics of geom_line are linetype, color and linewidth. It is good to note that linewidth does exactly the same thing as size and thus can be used interchangibly.

Example:

ggplot(mpg, aes(x = cty, y = hwy)) +
  geom_line(color = "purple")

ggplot(mpg, aes(x = cty, y = hwy)) +
  geom_line(color = "purple") +
  geom_point(color = "yellow", size = 1)

\(\cdot\) geom_smooth():

This geom function creates what we call a trend line or the best-fit line. A trend line aids the eye in seeing patterns in the data or understanding the relationship between X and Y variables. It is often used in combination to geom_point to help better visualization of the trends. Eventhough, the function requires a y variable, it generates its own y values using the provided y variable and uses that to plot the best fit line. In last 1/3rd of the course, we will use this function a lot.

Most commonly used arguments of geom_smooth in this course are method, formula and se.

• The method argument is used to specify the smoothing method, by default the smoothing method is automatically selected by R depending on the sample size. In this course we will use method as ‘lm’ for linear models. formula can also be used to specify the exact formula for your method of smoothing, for example, lm uses y~x and you can use formula = y~x.

• The se argument is the argument that displays the standard error (also known as the confidence interval) band of the best fit line. It uses values TRUE or FALSE depending on whether you want to display or not display the se band, the default being TRUE.

Example:

ggplot(mpg, aes(displ, hwy)) +
  geom_point() +
  geom_smooth()

ggplot(mpg, aes(displ, hwy)) +
  geom_point() +
  geom_smooth(method = "lm", se = F)

\(\cdot\) geom_col():

This geom function creates bar charts for discrete/categorical variables where the heights of the bars are created using values from a dataset. This geom treats each axis differently and, thus, can thus have two orientations. It uses one of the axis as a named variable or discrete variable and the other axis as a continuous variable.

Most commonly used aesthetic for geom_col is color and fill.

Example: Let’s say we wish to create a bar chart for the class variable in the mpg dataset that shows the number of cars or counts in each class on the y-axis.

ggplot(count(mpg, class), aes(class, n)) + # the count function counted the number of cars in each class. So how many compact cars, how many 2seater, etc. We will learn more about count in next chapter notes.
  geom_col(color = "purple", fill = "yellow")

We will see later that, if the variable in the y axis is the count there’s a one variable plot that is more convenient to use than geom_col. But the geom_col is more flexible as it gives you choices other than count in the other axis.

Now we move onto the one-variable plots.

\(\cdot\) geom_bar():

This geom function creates bar charts for discrete/categorical variables where the height of the bar is proportional to the number of observation in each group. geom_bar is equivalent to geom_col with the other variable being the count of the given categorical variable.

It uses the same aesthetics as geom_col. The x-axis is

ggplot(mpg, aes(class)) +
  geom_bar(
    color = "purple",
    fill = "yellow"
  )

Note, that the here geom_bar created the exact same graph as geom_col in the previous plot without the counts given to it. It also arranges the x-axis alphabetically by default. If you wish to arrange the chart according to the height of the bars, you will need to use reorder. You can see an example of that in your discussion assignment.

\(\cdot\) geom_histogram():

geom_histogram helps visualise the distribution of a single continuous variable by dividing the x (or y) axis into bins and counting the number of observations in each bin. Each column shows the frequency in the given interval.

Most commonly used aesthetics for geom_histogram are the color and fill. Choosing a reasonable binning scheme is subjective but very important part of creating a histogram. Thus, binwidth, bins, center and boundary are some of the most commonly used arguments. All of these arguments take in a single number.

• binwidth is how wide you want each interval to be.

• bins is how many bins you want to end up with. You cannot specify both binwidth and bins.

• center allows you to declare you want a bin centered around a specific number.

• boundary allows you to declare you want a certain boundary between two bins. You cannot specify both center and boundary.

ggplot(mpg, aes(x = displ)) + 
  geom_histogram(color = "purple", fill = "yellow")

From the way the x-axis is labeled, we can’t tell exactly how wide each bin is, let alone what the two endpoints are. So, let’s say we need the bins from 0-0.5, 0.5-1, 1-1.5 and so one, we could use binwidth to be 1 and boundary to be 0 (or center to be 0.5, boundary and center can be used alternatively).

ggplot(mpg, aes(x = displ)) + 
  geom_histogram(color = "purple", fill = "yellow", binwidth = 0.5, boundary = 0)

\(\cdot\) geom_density():

geom_density computes and draws kernel density estimates, which is a smoothed version of the histogram. This is a useful alternative to the histogram for continuous data that comes from an underlying continuous distribution.

Most commonly used aesthetics for geom_density are color, fill, linewidth or size, linetype and alpha.

ggplot(mpg, aes(displ)) +
  geom_density(
    color = "purple",
    fill = "yellow"
  )

Since, geom_density emphasizes the general trend in the data and geom_histogram shows the frequency of the raw data. Sometimes it is useful to see them both in the same plot. We can overlay the density plot on a histogram or vice versa and tuning down the opacity of one on the top. But we face a problem, when plot both of these in one plot, due to the significant difference on the scale of their y-axis, density plot only appears as a line on the bottom.

ggplot(mpg, aes(displ)) +
  geom_histogram(fill = "skyblue1") +
  geom_density(color = "red", fill = "pink", alpha = 0.3)

To handle that, we can set the y variable of the histogram as after_stat(density), which will scale down the axis of the histogram to fit that to the density.

ggplot(mpg, aes(displ)) +
  geom_histogram(
    aes(y = after_stat(density)), # This line shrinks the histogram's height to be on the same scale as geom_density
    fill = "skyblue1"
  ) +
  geom_density(
    color = "red",
    fill = "pink",
    alpha = 0.3, # alpha takes value between 0 and 1. The closer the value is to 0, the more transparent the plot is. The density plot here is "on top".
  )

\(\cdot\) geom_boxplot():

geom_boxplot compactly displays the distribution of a continuous variable. It visualises five quantities and all “outlying” points individually. The quantities are: - The minimum (or in the presence of outliers, the smallest data value bigger than Q1-1.5IQR) - The first quartile, Q1 - The second quartile or median, Q2 - The third quartile, Q3 - The maximum (or in the presence of outliers, the largest value less than Q3+1.5IQR)

Here, Q1 is the 25th percentile, Q2 is the 50th percentile and Q3 is the 75th percentile. IQR is the interquartile range or Q3-Q1 also represented by the box in the boxplot. Note: The Xth percentile is the value at which X% of the data is below it.

A boxplot is fundamentally different from the other one-variable geoms.

ggplot(mpg, aes(displ)) +
  geom_boxplot(fill = "orange")

It is important to note that the y-axis in the boxplot does not represent anything meaningful like the other one-variable geoms. Here all the relevant information is obtained from the x-axis.

Using a variable aesthetic for color or fill

Let’s try making some of the plots we built above and using a variable aesthetic for color or fill. This is just a demonstration of how variable aesthetics other than x and y-axis can be applied to plots. You can explore and see how other aesthetics such as size, shape, etc will be affected.

Since it’s a variable aesthetic, we will be providing it inside the aes(). Let’s revisit one of the plots we have made before. Say, for the mpg dataset, you want to make a scatterplot for hwy vs cty plot and color the points according to the class of car.

ggplot(mpg, aes(x = cty, y = hwy, color = class)) +
  geom_point(size = 1)

Let’s say you want to understand the relationship between hwy and displ for each fuel type. So we will build a scatterplot for hwy vs displ and color depending on fuel type. This should create 5 colors for the 5 fuel types.

ggplot(mpg, aes(x = displ, y = hwy, color = fl)) +
  geom_point() +
  geom_smooth()

You can see that it has colored the points and the trend lines according to the fuel types. We don’t see a trendline for fuel type ‘c’ and ‘d’ since we don’t have enough data to build our model on.

Now let’s say we want to understand the distribution of the type of drive and make a bar chart for that. Also, color of each bar depending on the type of drive. Before we can make a bar graph for just automatic and manual transmission, we need to modify all the auto() entries to just auto and all the manual() into just manual.

mpg_1 = mpg #save the dataset from the package to your environment
mpg_1$trans = mpg_1$trans %>% str_extract("auto") #the str_extract extracts the word "auto" from the rows of trans column of mpg_1 and if the row doesn't have "auto" it replaces it with NA 
mpg_1$trans[which(is.na(str_extract(mpg_1$trans, "auto")))] = "manual" # here we replace those NAs with manual
ggplot(mpg_1, aes(y = trans, color = trans, fill = trans)) +
  geom_bar()

Now let’s make boxplots from the displ depending on the cyl. Here, we have to be a little cautious, since the cyl variable is numeric, R doesn’t consider it categorical expects all real numbers between 4 and 8, and thus can’t produce seperate graph for individual cylinders. Thus, we need to make it categorical, by using the function as.factor().

ggplot(mpg, aes(displ, fill = as.factor(cyl))) +
  geom_boxplot()

Play with some more of the ideas like making histograms and density plots for some variable and supplying a second variable as color or fill. We will now move to some exercises and then introduce ways to facet and customizations for plots such as labels, themes, titles, scales and more

Everything covered after this point will not be tested on the Exams but will be used for HWs and DISs.

Moving forward we will just use the mendota dataset to demonstrate faceting and customizations.

\(\cdot\) facet_wrap():

Faceting with facet_wrap is a way to replicate a single plot within each subgroup defined by a categorical variable.

• When replicating a single plot, we reviewed in a previous exercise how to use color or fill to overlay separate marks for each subgroup on the same panel, as below.

ggplot(mendota, aes(x= duration, fill = century, linetype = century)) +
  geom_density(alpha = 0.3)

• However, we may also just want to split each onto its own plot. This is called faceting.

• The function facet_wrap requires one argument, facets; the variable by which you want to split the plot. One panel will be generated for each category of that variable.

  • facet_wrap requires you to surround this variable with the vars() function, like in the example below.
  • Unfortunately, this is just something that you have to memorize. If you do not use vars(), it will say object 'century' not found, or whatever variable you used.

ggplot(mendota, aes(x = duration, fill = century)) +
  geom_density() +
  facet_wrap(facets = vars(century))

\(\cdot\) facet_grid()

You can also facet by two variables with facet_grid, which requires you to specify the rows variable and cols variable with vars(). This is most useful when you have two variables for which every combination exists in the data. For example, faceting by decade and century doesn’t help much, because each decade only appears in one century.

ggplot(mendota, aes(x = duration)) +
  geom_density() +
  facet_grid(rows = vars(decade), cols = vars(century))

• Perhaps a more effective choice to communicate the same information as above would be to facet by decade and fill by century.

ggplot(mendota, aes(x = duration, fill = century)) +
  geom_density() +
  facet_grid(rows = vars(decade))

• Consider a column leap_year which identifies if year1 for each winter was a leap year.
  • Code to create this column is included in the .Rmd but suppressed in the knitted file.

## # A tibble: 6 × 2
##   year1 leap_year
##   <dbl> <lgl>    
## 1  1855 FALSE    
## 2  1856 TRUE     
## 3  1857 FALSE    
## 4  1858 FALSE    
## 5  1859 FALSE    
## 6  1860 TRUE

• Leap years have occurred in every century; so it makes sense to facet by both century and leap_year.

ggplot(mendota, aes(x = duration)) +
  geom_density() +
  facet_grid(rows = vars(century), cols = vars(leap_year))

\(\cdot\) Adding refernce lines to your plots

Adding a reference line to the graph sometimes makes it easier to understand some context. There are three functions that can be used to do so: geom_vline(), geom_hline() and geom_abline().

• geom_vline() creates a vertical line. It requires the xintercept that you provide in a vector, which controls the horizontal position of the line.

• geom_hline() creates a horizontal line. It requires the yintercept that you provide in a vector, which controls the vertical position of the line.

• geom_abline() creates a line with some slope and intercept (y-intercept). It requires the slope and intercept, which controls the placement of the line on the plot.

Unlike most other geoms, these geoms do not inherit aesthetics from the plot default, because they do not understand x and y aesthetics which are commonly set in the plot. They also do not affect the x and y scales. Thus, they additionally accept useful constant aesthetics like size, color, and linetype.

Example: Annotating where the mean is on a histogram; the value of the mean needs to be calculated before it is provided as the xintercept.

meanDuration = mean(mendota$duration)

ggplot(mendota, aes(duration)) +
  geom_histogram(
    color = "steelblue4",
    fill = "skyblue1"
  ) +
  geom_vline(xintercept = meanDuration,
             size = 0.5, linetype = "dashed", color = "red") 

# Note: even though xintercept = meanDuration technically has a variable in it, that variable just contains one single number. It is not a column in the dataframe mendota. Therefore, this is a constant aesthetic and does not require aes().

# A reminder: geom_vline() should come AFTER geom_histogram! What would happen if we put geom_vline() before the geom_histogram?

• You can also give a vector of values as the xintercept.

usefulValues = meanDuration + c(-3, -2, -1, 0, 1, 2, 3) * sd(mendota$duration)
# A vector with length six; these values are meaningful statistically, we'll learn why in the second half of the course
usefulValues

## [1]  42.29360  62.40162  82.50963 102.61765 122.72566 142.83368 162.94169

ggplot(mendota, aes(duration)) +
  geom_histogram(
    color = "steelblue4",
    fill = "skyblue1"
  ) +
  geom_vline(xintercept = usefulValues,
             size = 0.5, color = "darkblue") 

• And finally, an example of geom_hline. Remember those outliers that geom_boxplot identified? We can identify them on the scatterplot too!

iqr = IQR(mendota$duration)
firstQuartile = quantile(mendota$duration, 0.25)
thirdQuartile = quantile(mendota$duration, 0.75)

ggplot(mendota, aes(x = year1, y = duration)) +
  geom_point() +
  geom_hline(
    yintercept = c(firstQuartile - iqr, thirdQuartile + iqr)
    )

\(\cdot\) Scales:

Editing graphical properties of the axes is done with the family of scale_x_* and scale_y_* commands.

• The asterisk specifies the type of variable on that axis. For example, continuous for variables like duration (which can take on any numeric value in a given range), or discrete for variables like century (which only take on one of a finite set of categories).

• We will most commonly use:

  • scale_x_continuous()
  • scale_y_continuous()
  • scale_x_discrete()
  • scale_y_discrete()

• Just like geoms, there are too many examples of scale functions to go over in one lecture; we will see many over the course of the class.

• Helpful arguments you can pass into scale functions include:

  • breaks, a vector of locations to draw grid lines and labels at.
  • labels, a vector of names to use as the label of each break-point.
  • limits, a vector of two numbers specifying the left and right limit of how wide/tall you want the plot to be
  • trans, standing for “transformation”, which allows you to do some numeric transformation of the axis; including “reverse”, “sqrt”, and “log”.

# Notice ggplot's default x-axis choices
ggplot(mendota, aes(duration)) +
  geom_histogram(
    color = "steelblue4",
    fill = "skyblue1"
  )

ggplot(mendota, aes(duration)) +
  geom_histogram(
    color = "steelblue4",
    fill = "skyblue1"
  ) +
  scale_x_continuous(
    breaks = c(30, 90, 150),
    labels = c("1 month", "3 months", "5 months"),
    limits = c(15, 165),
    minor_breaks = NULL, # This specifies not to draw any vertical axis lines between the labeled points; not necessarily something you have to memorize, just an example of how far you can customize!
  )

ggplot(mendota, aes(duration)) +
  geom_histogram(
    color = "steelblue4",
    fill = "skyblue1"
  ) +
  scale_x_continuous(
    breaks = c(30, 90, 150),
    labels = c("1 month", "3 months", "5 months"),
    limits = c(-100, 300),
    minor_breaks = NULL
  ) +
  # Can you figure out what this addition is doing to the y-axis?
  scale_y_continuous(
    expand = expansion(mult = c(0,0.1)),
    limits = c(-10, 100)
  )

\(\cdot\) Color Scales:

When color is mapped to a variable aesthetic, you can use the viridis color scales for accessible preset options, or use the manual functions to set a custom color scale.

Recall the following plot from a previous exercise:

ggplot(mendota, aes(x= duration, fill = century)) +
  geom_density(alpha = 0.3)

ggplot’s default color schemes can be hard to distinguish for people with common forms of color blindness. The “viridis” color scales are designed to remedy this. Depending on whether your variable is continuous (c) or discrete (d), and whether you used color or fill as the aesthetic, you can use one of the following four commands: - scale_color_viridis_c() - scale_color_viridis_d() - scale_fill_viridis_c() - scale_fill_viridis_d()

For example, in the plot above, we use fill as the aesthetic controlling color, with century a discrete/categorical variable, so we use scale_fill_viridis_d().

Examples:

ggplot(mendota, aes(x= duration, fill = century)) +
  geom_density(alpha = 0.3) +
  scale_fill_viridis_d()

ggplot(mendota, aes(x= duration, fill = century)) +
  geom_density(alpha = 0.3) +
  scale_fill_viridis_d(option = "inferno")

Alternatively, you might have a custom color scheme in mind. scale_color_manual and scale_fill_manual exist to help you; the values argument accept a vector of pairs, where you map values of the categorical variable to colors.

ggplot(mendota, aes(x= duration, fill = century)) +
  geom_density(alpha = 0.3) +
  scale_fill_manual(
    values = c("19th" = "dodgerblue", "20th" = "peachpuff", "21st" = "mediumorchid")
    )

There are many options within viridis, see here (scroll a little down) for more details.

\(\cdot\) Plot Labels:

All plot labeling can be done with the labs() (standing for labels) function.

labs() can be used to add a title, subtitle, and caption; see placement examples below. It can also be used to adjust the axes labels and legend titles. The legend title can be changed through labs() by using the name of whatever aesthetic you used to create the legend.

For example, in this plot we create the legend with fill = century, so the legend title is adjusted with fill = "intended legend title".

densityPlot = ggplot(mendota, aes(x= duration, fill = century)) +
  labs(
    title = "Distribution of Freeze Duration by Century",
    subtitle = "Lake Mendota, 1855-2023",
    caption = "STAT 240",
    
    x = "Duration (in days)",
    y = "Density",
    fill = "Century" # If you created your legend with the size aesthetic, this would be size = "legend title", or color would be color = "legend title", et cetera
  ) +
  geom_density(alpha = 0.3) +
  scale_fill_manual(
    values = c("19th" = "dodgerblue", "20th" = "peachpuff", "21st" = "mediumorchid")
    )

densityPlot

\(\cdot\) Themes:

ggplot2 comes with many built-in themes to improve the appearance of the graph over the default theme, such as theme_minimal().

densityPlot +
  theme_minimal()

densityPlot +
  theme_classic()

This link contains a complete list of themes.

\(\cdot\) Shortcut Functions:

The “general” form of all the customization functions are discussed above. Many of the more common tasks have shortcut functions; they are useful if you only need to make one change. Note: The general form because they can accomplish everything these shortcuts can do and more, and you have less functions to memorize.

Examples of shortcut functions include:

• xlim(c(a, b)) is the same as scale_x_continuous(limits = c(a, b)), and similarly for y.

• scale_x_reverse() is the same as scale_x_continuous(trans = "reverse"), and similarly for y.

• ggtitle("my title") is the same as labs(title = "my title").

• xlab("x axis title") is the same as labs(x = "x axis title"), and similarly ylab() can be used for y.

Another EXERCISE: This uses all the topics we have covered in the lecture note.

Interpreting a Faceted Plot: Consider the graph below and choose from the given options to correctly interpret the plot:

ggplot(mendota, aes(x = duration)) +
  geom_density() +
  facet_grid(rows = vars(century), cols = vars(leap_year))

• The top left panel shows the distribution of duration among (leap years/non-leap years) in the (19th/20th/21st) century.

• The bottom right panel shows the distribution of duration among (leap years/non-leap years) in the (19th/20th/21st) century.

• We don’t expect there to be a difference in average duration between non-leap years and leap years. This is illustrated by the fact that each (row of panels/column of panels) has roughly the same center across each of its panels.

• We do expect there to be a difference in average duration across centuries. This is illustrated by the fact that each (row of panels/column of panels) has different centers across each of its panels.

Technical takeaway: The subgroup represented in an individual faceted panel can be defined by one OR two variables; the faceting commands do a decent but not perfect job of labeling them.

Philosophical takeaway: Faceting is another valuable tool for showing two-variable relationships. It is especially helpful when we have too many subgroups to overlay on a single panel.

• Philosophical takeaway continued: Notice how difficult it is to encode leap_year AND century with just aesthetics.

ggplot(mendota, aes(x = duration, fill = century, linetype = leap_year)) +
  geom_density(alpha = 0.5, size = 1)

In the mpg dataset use the variables cty, hwy displ, drv and year to recreate the graph below. Use all the concepts we have learned so far.

image

ggplot(mpg, aes(cty, hwy, col = displ)) +
  geom_point() +
  geom_smooth(method = "lm") +
  facet_grid(rows = vars(year), cols = vars(drv)) +
  theme_minimal() +
  scale_color_viridis_c()

---